Loops HouseRunLatch

NEAR HACKATHON
Links
Team
1 member- DAOwner
davee0x
Overview
RunLatch is a project-based purchasing and accountability system for businesses that use AI agents to complete client work.
As agents become more capable of working independently, businesses need a way to let them purchase the resources they need without giving them unrestricted access to funds. RunLatch addresses this by allowing an owner to create a Client Work Capsule with a defined budget, approved vendors, purchasing limits, and project-specific rules.
When an agent requests a purchase, RunLatch's deterministic Gate checks the request against the registered quote and the project's permissions. It can identify altered purchase details, requests that exceed the approved limits, and unexpected recurring commitments, blocking or escalating requests that do not meet the rules.
RunLatch also connects purchasing decisions to what happens afterward. It keeps authorisation, payment status, provisioning evidence, objective validation checks, and human acceptance as distinct records, giving owners a clearer picture of what was approved, what was spent, and what evidence exists that the expected resource was delivered.
NEAR provides the smart-contract environment for enforcing and testing these purchasing rules. We implemented the contract logic in Rust and tested it using NEAR Sandbox. The current demo combines clearly labelled browser fixtures with local Sandbox evidence; it does not claim a completed public testnet purchase.
Our goal is simple: give AI agents the freedom to get work done while keeping their spending accountable to the people and projects they serve.